Project Name

09.04.20XX
─

Your Name
Your Company
123 Your Street
Your City, ST 12345
 1

Overview
Lorem ipsum dolor sit amet, consectetuer adipiscing elit, sed diam nonummy nibh
euismod tincidunt ut laoreet dolore magna aliquam erat volutpat. Ut wisi enim ad minim
veniam, quis nostrud exerci tation ullamcorper.

Goals
1. Lorem ipsum dolor sit amet, consectetuer adipiscing elit

2. Sed diam nonummy nibh euismod tincidunt ut laoreet dolore magna aliquam erat
volutpat.

Specifications
Nam liber tempor cum soluta nobis eleifend option congue nihil imperdiet doming id quod
mazim placerat facer possim assum. Typi non habent claritatem insitam; est usus legentis
in iis qui facit eorum claritatem. Investigationes demonstraverunt lectores legere me lius
quod ii legunt saepius.

Lorem Ipsum
Duis autem vel eum iriure dolor in hendrerit in vulputate velit esse molestie consequat, vel
illum dolore eu feugiat nulla facilisis at vero eros et accumsan.

Milestones

I. Lorem ipsum
Lorem ipsum dolor sit amet, consectetuer adipiscing elit, sed diam nonummy nibh
euismod tincidunt ut laoreet dolore magna aliquam erat volutpat.

II. Dolor sit amet

Lorem ipsum dolor sit amet, consectetuer adipiscing elit, sed diam nonummy nibh
euismod tincidunt ut laoreet dolore magna aliquam erat volutpat.
 2

Estimate the regression of ln_time_learn on the explanatory variables as mentioned

above. Report the coefficient estimates and the p-values in a table. Interpret and explain
the results considering both the significant and insignificant coefficients. In comparison
to the model with age and age squared, which of the two models will you choose and
how?

AGE GROUPS(ref COEFFICIEN P-VALUE INTERPRETATION

18-21) T

22-25yrs -0.297*** 0.000 Younger age groups spend

significantly less time learning than the
reference (18-21yrs).

26-29 yrs -0.658*** 0.000 The negative effect strengthens with

age.

Sector (Ref: Rural)

Urban 0.0347*** 0.000 Urban dwellers spend ~3.5% more

time learning.

Gender(Ref: Male)

Female -0.0524*** 0.000 Females spend ~5.2% less time

learning than males.

Social
Group(Ref:General)

ST -0.00836 0.470 Insignificant Effect

SC -0.00525 0.576 Insignificant Effect

OBC -0.00616 0.397 Insignificant Effect

Religion(Ref:Hindu)

Muslim -0.0172 0.080

 3

Christian 0.00869 0.543 Insignificant effect

Others -0.0487** 0.001 Significant negative effect

Log MPCE 0.0530*** 0.000 Higher income → more

learning time.

Household Size -0.0102*** 0.000 Larger households → slightly

less learning time.

Constant 5.683*** 0.000 Baseline learning time (when all

controls = 0).

The regression equation is as follows

ln(time_learn) = β 0+ β 1Age 18-21 + β 2Age 22-25 + β 3Age 26-29+ β 4 urban+ β 5female+ β 6ST
+ β 7SC+ β 8OBC+ β 9Muslim+ β 10Christian+ β 11OtherReligion + β 12ln MPCE+ β 13HH_Size+ μ

With proper coefficients

ln(time_learn)=5.683−0.101⋅Age 18-21 −0.297⋅Age 22-25
−0.658⋅Age 26-29
+0.035⋅Urban−0.052⋅Female−0.008⋅ST−0.005⋅SC−0.006⋅OBC
−0.017⋅Muslim+0.009⋅Christian−0.049⋅OtherReligion+0.053⋅ln(
MPCE)−0.010⋅HH_Size+u

VARIABLES COEFFICIENT P-VALUE INTERPRETATION

Age 0.0988*** 0.000 For each additional year of age,

learning time initially increases
by ~9.9%, holding other factors
constant.

Age
2
-0.0036*** 0.000 The positive effect of age
diminishes over time—learning
time increases at a decreasing
rate and eventually declines.
 4

Learning time is highest around

12.4 years (though this may be
outside the sample range,
suggesting a declining trend in
the observed age groups).

Sector(Urban) 0.0377*** 0.000 Urban residents spend ~3.8%

more time learning than rural
residents, controlling for other
factors.

Gender(Female) -0.0527*** 0.000 Females spend ~5.3% less time

learning than males, all else
equal.

Social Groups

ST -0.008 0.488 No statistically significant

differences in learning time
across these groups compared
to the reference (likely upper
castes).

SC -0.006 0.527 No statistically significant

differences in learning time
across these groups compared
to the reference (likely upper
castes).

OBC 0.006 0.428 No statistically significant

differences in learning time
across these groups compared
to the reference (likely upper
castes).
 5

Religion(Ref:Hindu)

Muslim -0.021 0.028 Muslim individuals spend ~2.1%

less time learning than Hindus.

Christian 0.010 0.471 No significant difference from

Hindus.

Others -0.049 0.001 Individuals from other religions

(e.g., Sikh, Jain, Buddhist) spend
~4.9% less time learning than
Hindus.

Log MPCE 0.058 0.000 A 1% increase in household

income leads to a 0.058%
increase in learning time.

Household Size -0.011 0.000 Each additional household

member reduces learning time
by ~1.1%, likely due to resource
dilution.

Constant 4.997 0.000 The predicted

ln⁡(time_learn)
ln(time_learn) for a rural, male,
Hindu, general-caste individual

with:

● Age = 0 (theoretical, not

realistic),
● ln⁡(MPCE)=0
● ln(MPCE)=0 (income =
₹1),
● Household size = 0.

The Regression Equation is as follows

 6

With the regression coefficients

ln(time_learn)=4.997+0.099⋅Age −0.004⋅ Age2
+0.038⋅Urban−0.053⋅Female−0.008⋅ST−0.006⋅SC−0.006⋅OBC−0.021⋅Musli
m+0.010⋅Christian−0.049⋅OtherReligion+0.058⋅ln(MPCE)−0.011⋅HH_Size+u

Model 2 (Quadratic Age) is statistically and theoretically superior to Model 1 .

Because of the following reasons

1. Flexibility in Functional Form

● Model 2 (Quadratic Age) allows for a smooth, non-linear relationship between

age and learning time.
○ Captures diminishing returns (learning time first increases, then
decreases).
○ More efficient than Model 1 (categorical dummies), which imposes a
step-function (less realistic).
● Wooldridge’s Rule:
"When a continuous variable (e.g., age) has a non-linear effect, quadratic terms are
preferred over arbitrary categorical splits."

2. Higher Explanatory Power (R²)

● Model 2 (Quadratic):
○ R² = 0.0648
○ Adj. R² = 0.0645
● Model 1 (Categorical):
○ R² = 0.0587
○ Adj. R² = 0.0584
● Conclusion: Model 2 explains more variation in learning time.

3. Fewer Parameters (Parsimony Principle)

 7

● Model 2 uses 2 coefficients (Age + Age²) to model non-linearity.

● Model 1 uses 3 dummy variables (Age 18-21, Age 22-25, Age 26-29).
● Wooldridge’s Emphasis:
"Prefer models with fewer parameters when possible, unless theory justifies
complexity."

4. Statistical Significance of Quadratic Term

● Age² coefficient = -0.004 (p < 0.001) is highly significant.

○ Confirms non-linearity is empirically important.
● Model 1 ignores this curvature, risking misspecification bias.

5. Marginal Effects Interpretation (Wooldridge, Ch. 6)

● Model 2 allows calculating:

○ Turning point (peak learning time at Age = 12.4 years).
○ Marginal effect of age:

○
■ Example: At Age = 20 → Marginal effect = 0.099 -
0.008×20 = -0.061 (learning time declines by 6.1% per year).
● Model 1 only gives discrete jumps between age groups (less precise).

(2) Estimate the model in (1) and now use females as the reference category.
Compare the results by reporting the values for estimated coefficients, standard
error, t-stats with column headings as Models 1 and 2a respectively and indicate
the changes if any in the ‘gender’ dummy variable and the intercept for the
estimated values, the standard error, t-statistics, R2 . Use the simpler version in
Q3 in the dummy variables problem set to explain the differences/no differences
that you observe in these results.
 8

Variable Model1(Ref: Model 2 (Ref: Interpretation

Male) Female)
Gender(Female) -0.52 0.52 As compared to males
female section of the
study on an average
spends 5.33% less time
on learning
Gender(Male) 0 0.52 As compared to
females male section
of the study on an
average spends 5.33%
more time on learning
Intercept 5.68 5.63 The change in the
dummy variable is
exactly offset by the
change in the intercept
value.
T-statistics -8.70 8.70 The magnitude
remains the same
however we see a flip
in sign , as the dummy
variable measures the
effect relative to the
base category. In other
words, belonging to
Male group would
increase the time spent
on learning coefficient
value by 5.2% and
decrease it by the same
amount if belonging to
 9

female group
Standard errors 0.006 0.006 There is no change in
the SE as only the base
category of the dummy
variable.
R² 0.058 0.058 R squared measures
the proportion of
variance explained by
the [Link] both
models explain the
same exact variation
and the set of
regressors is the
same ,R^2 is
unchanged.

(3) Estimate the model in (1) and instead of the dummy variables for the social
groups use the variable social_groups as it is. Mention the value of this
estimated coefficient and its standard error and explain what is the difference in
the results and why you will not prefer to use the social group variable in this
manner.
Variable Coefficient Standard T statistics P value
Error
social_grp 0.0009 0.001 0.94 0.34

Using social_group as a continuous variable gives us a small, statistically

insignificant coefficient (p = 0.34). For each one-unit increase in the value of
social_group, the log of time to learn decreases by about 0.0009, which is
negligible.
In our earlier models (Model 1 and 2), social_group was entered as categorical
dummies: ST (Scheduled Tribe) ,SC (Scheduled Caste) ,OBC (Other Backward
Class) ,General (used as the reference category)
 10

Each group has no meaningful numeric relationship with the others. The
categories are qualitative, not quantitative. Using i.social_group creates
dummies and treats each category [Link] allows the regression to
estimate a separate effect for each group without assuming any order or
[Link] gives more accurate and meaningful interpretation, like:
"Compared to the General category, students from ST take 0.0083 units
more/less time to learn, etc."
So using social_group as a numeric variable assumes linearity — that the
difference between, say, ST (1) and SC (2) is the same as between SC (2) and
OBC (3), which is not valid.

4) Estimate (1) to include the interaction of the gender dummy with the age_group
dummy with all other explanatory variables as it is in the model. (Use the xi: regress
command in STATA to generate the interaction dummies appropriately.) Present the
results for model (1) and this model (4) in the same table and include the coefficients
(with * to reflect the significance) and only the p-values. Interpret the results by giving
and explanation on the advantage that you have with this model to explain the gender
differences in (log) learning time across age groups.

This analysis compares Model 1 (baseline with categorical age groups) and Model 4
(with gender-age interactions) to assess how gender differences in learning time vary
across age groups.

Key Regression Results

1. Model 1 (Baseline Model)

● Gender Effect (Female): -0.052 (p < 0.001)

○ Females spend 5.2% less time learning than males on average, holding
age and other factors constant.
● Age Effects:
○ 18-24 yrs: -0.101 (p < 0.001)
 11

○ 26-29 yrs: -0.297 (p < 0.001)

○ 30+yrs: -0.658 (p < 0.001)
○ Interpretation: Learning time declines sharply with age.

2. Model 4 (Gender × Age Interaction Model)

● Gender Effect (Female): 0.006 (p = 0.500, insignificant)

○ The baseline gender gap disappears when interactions are added.
● Interaction Terms:
○ Female × 18-24 yrs: -0.068 (p < 0.001)
○ Female × 26-29 yrs: -0.119 (p < 0.001)
○ Female × 30+yrs: -0.133 (p < 0.001)
○ Interpretation: The female disadvantage grows with age:
■ Females aged 18-24 spend 6.8% less time learning than males of
the same age.
■ Females aged 26-29 spend 11.9% less time.
■ Females aged 30+ spend 13.3% less time.

Advantages of Model 4 (Wooldridge’s Framework)

1. Captures Heterogeneous Effects

● Model 1 assumes a constant gender gap (-5.2%) across all ages, which is overly
simplistic.
● Model 4 reveals that the gender gap widens with age, aligning with Wooldridge’s
emphasis on testing interactions when subgroup differences are plausible (Ch.
7).

2. Avoids Omitted Variable Bias

● If gender effects vary by age (as they do here), failing to include interactions
biases the baseline gender coefficient (Wooldridge, Ch. 7).
○ In Model 1, the -5.2% female effect is an average that masks critical age-
specific trends.

3. Policy Relevance

● Model 4 identifies which age groups need targeted interventions:

○ The widening gap for older females suggests barriers like caregiving
duties or labor market pressures.
○ Model 1 would overlook these nuances.
 12

Why Model 4 is Preferable (Statistical & Theoretical Reasons)

1. Higher Precision: While both models have similar R² (~0.06), Model 4 provides
more actionable insights.
2. Theoretical Soundness: Wooldridge advises:
"Always include interaction terms when theory suggests effect heterogeneity." (Ch.
7)
3. Statistical Significance: All interaction terms are highly significant (p < 0.001),
confirming age modifies gender effects.

Conclusion

Model 4 (with interactions) is superior because it:

1. Reveals age-specific gender disparities missed by Model 1.

2. Follows Wooldridge’s guidelines for flexible specification.
3. Informs targeted policies (e.g., support for older female learners).

Ques 5

Table for Q5(a): Regression Results for Males and Females

(1) (2)

VARIABLES Males Females

ln_time_learn . .

(.) (.)

. .

2.age_grp -0.07*** -0.14***

 13

(0.01) (0.01)

-8.04 -13.56

3.age_grp -0.24*** -0.38***

(0.01) (0.01)

-20.09 -25.77

4.age_grp -0.52*** -0.86***

(0.02) (0.03)

-26.66 -33.79

[Link] 0.04*** 0.03***

(0.01) (0.01)

4.77 2.83

1.social_group -0.04** 0.02

(0.02) (0.02)

-2.36 1.30

2.social_group 0.00 -0.02

(0.01) (0.01)

0.26 -1.18

3.social_group 0.00 -0.02*

(0.01) (0.01)

0.45 -1.77

[Link] -0.01 -0.02

(0.01) (0.01)

-1.09 -1.42
 14

[Link] 0.02 -0.01

(0.02) (0.02)

1.23 -0.31

[Link] -0.07*** -0.02

(0.02) (0.02)

-3.69 -0.67

lnmpce 0.04*** 0.07***

(0.01) (0.01)

5.08 7.87

hhsize -0.01*** -0.01**

(0.00) (0.00)

-5.33 -2.53

Constant 5.77*** 5.51***

(0.06) (0.08)

90.28 72.02

Observations 23,008 17,911

R-squared 0.04 0.08

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

 15

Use the gender dummy interacted model by retaining only the variables that are
significant as in model 5(b).
(6) Apply the White’s test for heteroscedasticity based on the regular and
special case of this test. Present the results of the respective auxiliary
regressions in a table with the test-statistic and the tabulated values included in
it. Comment on your findings for heteroscedasticity.

This model includes significant interactions of gender with age group, social
group, and lnmpce.
Main Regression Equation:
ln⁡(time_learn)=β0+β1 age_grp+β2 gender+β3 (age_grp×gender)+β4
social_group+β5 (social_group×gender)+β6 ln⁡(mpce)+β7 (ln⁡(mpce)×gender)+u

Where:

● ln_time_learn = log of time taken to learn

● age_grp, social_group, gender = categorical variables

● lnmpce = log of monthly per capita consumption expenditure

● u = error term

This model includes interaction terms like:

● 26-29yrs * Female

● ST * Female

● lnmpce * Female

Auxiliary Regression Equation (for White’s Test)

In White's test, we regress the squared residuals u^2\hat{u}^2u^2 on:

 16

● All the variables from the main model

● Their squares (like ln⁡(mpce)2\ln(\text{mpce})^2ln(mpce)2) and/or

interaction terms to capture non-linearity and cross-effects

Auxiliary Regression Equation (Regular Case):

m û=α0+α1 age_grp+α2 gender+α3 (age_grp×gender)+α4 social_group+α5
(social_group×gender)+α6 ln⁡(mpce)+α7 (ln⁡(mpce)×gender)+α8 (ln⁡(mpce))2+e
Where:

● û2 = squared residuals from the main regression

● e = new error term for the auxiliary regression